Pyron has a built-in assistant you can ask for help in plain language. It reads the entries you already have access to, answers questions about them, and drafts changes — a new form, a report, an automation — that you review before anything takes effect.
Where the assistant lives
The assistant turns up in two places. Inside an editor — the form builder, the report editor, the flow-graph editor, or a dashboard widget editor — it sits in a panel beside what you are working on, so you can ask it to reshape that form, report or automation and watch the result take shape. Anywhere else in Pyron, you can open it as a chat panel to ask a question or start something new.
The panel collapses to a slim edge and reopens when you want it. Collapsing it keeps your conversation, so a reply that is still coming through is not lost.
From the app-wide assistant you can also begin from nothing: describe the form, report or automation you have in mind, and Pyron creates it and opens its editor with the assistant already alongside, ready to keep refining it.
What it can do
The assistant does two broad kinds of work: it answers questions about your data, and it drafts changes you can apply.
Ask a question and it looks through your entries to answer — matching both on the words you used and on meaning, so it can surface entries that are about what you asked even when they are worded differently (see Search). It can reply with a short table or a summary card built from what it found, not only a paragraph.
When you ask it to build or change something, it can:
- Put together or adjust a form, including the fields people fill in — see Building forms with the assistant.
- Draft a report template — see Drafting templates with the assistant.
- Set up an automation that runs on its own once you turn it on.
- Add or rework a widget on a dashboard.
What it acts on depends on where you opened it. Open it inside the form builder and it works on that form; open it beside a dashboard widget and it works on that widget.
Drafts you approve, never silent changes
The assistant never changes your work on its own. When it proposes a change, that change arrives as a draft: you see what it would do, and it takes effect only when you apply it.
In an editor, a draft appears as a notice above your work with the option to apply it. Read what it suggests, then apply it or leave it and carry on — and, as with any edit, the change is only kept once you save. If a draft is not what you wanted, ignore it and ask for a different one; the conversation keeps every draft it has offered, so you can go back and apply an earlier one.
A card the assistant shows you in chat is an answer, not a draft — there is nothing to apply. And because applying a draft to an automation replaces the whole diagram at once, that one is applied as a deliberate step, with an undo if you change your mind.
A draft is a suggestion until you apply it, and applying it still leaves the normal save step in your hands. You can always read a change in full before it becomes part of your work.
Privacy and guardrails
The assistant can only see what you can see. It reads your data through the same access rules as the rest of Pyron — your role and anything shared with you — so two people asking the same question can get different answers depending on which parts of the directory they can open. Asking the assistant never reveals anything you could not already reach yourself.
Every message to and from the assistant passes through guardrails. They screen for unsafe content, block secrets such as access keys, and mask sensitive numbers like payment-card numbers. Your administrator chooses how strict these rails are — see AI guardrails.
Which AI provider powers the assistant is also an administrator's choice. Pyron comes with a provider built in, and your organisation can connect its own instead. If your data needs to be processed in a particular region, the provider can be set so that requests stay there — see the AI provider settings. To keep usage predictable, an administrator can also cap how much the assistant is used each day (see AI budgets).